SheetAgent: Towards A Generalist Agent for Spreadsheet Reasoning and Manipulation via Large Language Models

2024-03-06

AI Summary

Real-world spreadsheet workflows often involve long-horizon, multi-step reasoning under ambiguous user intent—a challenge poorly addressed by existing methods. Method: We propose SheetAgent, an autonomous agent for spreadsheet manipulation, supported by the benchmark SheetRM. Its architecture comprises three synergistic modules—Planner, Informer, and Retriever—integrating structured chain-of-thought reasoning, dynamic context injection, semantic table retrieval, and decoupled action execution, augmented with a task-reflection mechanism to enhance robustness. Contribution/Results: This work is the first to systematically define and tackle long-horizon, ambiguity-driven multi-step spreadsheet operations. It introduces the first autonomous reasoning agent architecture specifically designed for spreadsheets. Empirical evaluation across multiple benchmarks demonstrates a 20–30% improvement in task success rate, significantly boosting operational accuracy and cross-task generalization—validating SheetAgent’s effectiveness and state-of-the-art performance in realistic office scenarios.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Spreadsheet manipulation is widely existing in most daily works and significantly improves working efficiency. Large language model (LLM) has been recently attempted for automatic spreadsheet manipulation but has not yet been investigated in complicated and realistic tasks where reasoning challenges exist (e.g., long horizon manipulation with multi-step reasoning and ambiguous requirements). To bridge the gap with the real-world requirements, we introduce $ extbf{SheetRM}$, a benchmark featuring long-horizon and multi-category tasks with reasoning-dependent manipulation caused by real-life challenges. To mitigate the above challenges, we further propose $ extbf{SheetAgent}$, a novel autonomous agent that utilizes the power of LLMs. SheetAgent consists of three collaborative modules: $ extit{Planner}$, $ extit{Informer}$, and $ extit{Retriever}$, achieving both advanced reasoning and accurate manipulation over spreadsheets without human interaction through iterative task reasoning and reflection. Extensive experiments demonstrate that SheetAgent delivers 20-30% pass rate improvements on multiple benchmarks over baselines, achieving enhanced precision in spreadsheet manipulation and demonstrating superior table reasoning abilities. More details and visualizations are available at https://sheetagent.github.io.

Problem

• Develops a generalist agent for spreadsheet reasoning and manipulation.
• Addresses challenges in long-horizon, multi-step spreadsheet tasks.
• Enhances precision and reasoning in spreadsheet manipulation using LLMs.

Innovation

• LLM-based autonomous agent for spreadsheets
• Three-module system: Planner, Informer, Retriever
• Iterative reasoning for precise spreadsheet manipulation